CN115687998A - Bearing state monitoring method based on signal open set incremental recognition model - Google Patents
Bearing state monitoring method based on signal open set incremental recognition model Download PDFInfo
- Publication number
- CN115687998A CN115687998A CN202211323575.8A CN202211323575A CN115687998A CN 115687998 A CN115687998 A CN 115687998A CN 202211323575 A CN202211323575 A CN 202211323575A CN 115687998 A CN115687998 A CN 115687998A
- Authority
- CN
- China
- Prior art keywords
- data
- bearing
- classes
- model
- open
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to the technical field of classification and identification, in particular to a bearing state monitoring method of an open set incremental identification model based on signals, which comprises the steps of preprocessing collected bearing data and extracting characteristics; reducing the dimension of the feature data processed in the first step according to the feature values by adopting a PCA algorithm to obtain a feature component after dimension reduction; and establishing an open set incremental model, identifying the state of the bearing, and detecting abnormal data. The invention provides an incremental open-set model applied to industrial bearing vibration signals, when new bearing data exist, an incremental learning method is adopted, the existing model is updated by using new knowledge, and an open-set recognition algorithm provides rejection options for a classifier so as to recognize new untrained target types.
Description
Technical Field
The invention relates to the technical field of classification and identification, in particular to a bearing state monitoring method based on an open set incremental identification model of signals.
Background
Machines in industry 4.0 are becoming more and more complex, which makes their vibration signal analysis a challenging task. Bearings are recognized as key components in industrial machinery, and therefore, mastering their vibration information can enhance the diagnostic process. The vibration signal of a defective bearing operating in a stationary condition may be considered an amplitude modulated waveform. This makes envelope analysis one of the most effective methods for bearing health monitoring under quiescent conditions.
Three frequency domain characteristics SPRO, SPRI and SPRR are extracted from vibration data in the Order-Based Identification of Bearing defect under Variable Speed Condition, a multi-core support vector machine (MSVM) classification model is established, and the state of a Bearing can be effectively identified; however, the industrial data is different day by day, and there is a possibility that a new fault signal occurs to cause the error recognition of the original model, and the model proposed by the literature cannot effectively recognize a new fault category, and needs to be retrained, which needs a lot of memory time.
Disclosure of Invention
Aiming at the defects of the existing algorithm, the invention provides an incremental open-set model applied to industrial bearing vibration signals, when new bearing data exist, an incremental learning method is adopted, the existing model is updated by using new knowledge, and the open-set recognition algorithm provides rejection options for a classifier so as to recognize new untrained target types.
The technical scheme adopted by the invention is as follows: a bearing state monitoring method based on an open set incremental recognition model of signals comprises the following steps:
step one, preprocessing collected bearing data and extracting characteristics;
further, preprocessing comprises dividing bearing vibration data and rotation speed data in a 2-second window;
further, the feature extraction includes performing mean, standard deviation, mode, maximum, minimum, skewness and kurtosis on the segmented data.
Step two, adopting a PCA algorithm to perform dimensionality reduction on the feature data processed in the step one according to the feature values to obtain a dimensionality-reduced feature component;
further, the method specifically comprises the following steps:
s21, setting n rows of d-dimensional data, and forming an n row and d column matrix X by the processed characteristic data according to columns; carrying out zero equalization on each column of X;
s22, solving an eigenvalue of the covariance matrix and a corresponding eigenvector; arranging the eigenvectors into a matrix according to the corresponding eigenvalues from large to small in rows, and taking the first k rows to form a matrix P; wherein Y = PX is data reduced to k dimensions.
Establishing an open set incremental model, identifying the state of the bearing, and detecting abnormal data;
further, the method specifically comprises the following steps:
firstly, training is started from K initial classes, wherein K is the number of classes of known classes, an initial model M0 is generated through an extreme value theory, bearing state data of a test sample are classified into K classes, and data of the rest classes are effectively rejected into unknown classes;
secondly, learning a new class, updating the model M0, and enabling the updated model to classify the bearing state data of the test sample into K +1 classes, and effectively rejecting the data of the rest classes which are not learned into unknown classes;
and finally, continuously adding unknown class data and identifying.
The invention has the beneficial effects that:
1. the open-set incremental model has better recognition effect on the new class samples than the existing method, only new data needs to be added for each training after the model is generated, and the model does not need to be trained again;
2. the method has good classification and identification effects on signal data in industries such as bearings, effectively reduces the cost of model training, has good effect on distinguishing new data, and has good application prospect in a man-machine integrated intelligent system.
Drawings
FIG. 1 is a flow chart of a method for monitoring a bearing condition based on an open set incremental signal identification model of the present invention;
FIG. 2 is a pareto chart corresponding to the dimensionality reduction of data of the present invention;
FIG. 3 is a F1 index plot in the open-set delta scenario of the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and examples, which are simplified schematic drawings and which illustrate only the basic structure of the invention and, therefore, only show the structures associated with the invention.
The Data set 'Bearing Vibration Data under Time-varying rotation Speed requirements' contains Vibration signals collected from bearings with different health Conditions under the condition of Time-varying rotation Speed, and 60 groups of Data are shared; for each set of data, there were two experimental settings: bearing health and transmission conditions. The health condition of the bearing includes (i) health, (ii) inner ring defect failure, (iii) outer ring defect failure, (iv) failure with ball defect, and (v) failure with composite defect of inner ring, outer ring and balls; the operating speed condition is as follows: (ii) increasing the speed, (ii) decreasing the speed, (iii) increasing then decreasing the speed, and (iv) decreasing then increasing the speed. Thus, there are 20 different cases; to ensure the authenticity of the data, 3 trials were collected per experimental setup, with a result of 60 total data sets. Each data set contains two channels: "Channel _1" is the vibration data measured by the accelerometer and "Channel _2" is the rotational speed data measured by the encoder. All of these data were sampled at a frequency of 200,000hz with a sampling duration of 10 seconds. The CPR (cycles per revolution) of the encoder is 1024.
As shown in fig. 1, a method for monitoring the condition of a bearing based on an open set incremental recognition model of a signal comprises the following steps:
step one, preprocessing and characteristic extraction are carried out on bearing data;
further, the extracted features include: mean, standard deviation, mode, maximum, minimum, skewness and kurtosis;
in the embodiment, a window size of 2S is selected for a bearing vibration signal, 300 groups of samples are generated, wherein 150 groups of test samples and 150 groups of training samples are respectively extracted, and 14 groups of characteristics are extracted, namely, a mean value, a standard deviation, a mode, a maximum value, a minimum value, skewness and kurtosis, according to vibration data measured by an accelerometer and rotation speed data measured by an encoder.
The health state of the bearing is shown in table 1, and there are five states.
TABLE 1 health status and abbreviations for bearings
Step two, performing feature dimension reduction by adopting a PCA algorithm;
PCA is adopted for feature dimension reduction, and is a common data analysis mode, is commonly used for dimension reduction of high-dimensional data, and can be used for extracting main feature components of the data;
the specific operation is as follows:
setting n rows of d-dimensional data, and forming an n row and d column matrix X by the original characteristic data according to columns; zero-averaging each column of X (representing an attribute), i.e., subtracting the mean of this column;
then, a covariance matrix is obtained
y is X zero equalized data; solving the eigenvalue of the covariance matrix and the corresponding eigenvector; arranging the eigenvectors into a matrix according to the corresponding eigenvalues from large to small in rows, and taking the first k rows to form a matrix P; wherein, Y = PX is data after dimensionality reduction to k dimensionality, and the dimensionality of Y is n × k.
In this embodiment, the PCA dimension reduction is adopted to reduce the original 14 sets of features to 5 sets of features, which are corresponding pareto graphs as shown in fig. 2, and it can be seen from the graph that the 5 features account for 90% of the total contribution (feature value), wherein the feature F14 accounts for 36% of the total contribution, so that F14, F11, F5, F8, and F12 are selected as final features; F1-F7 are the mean, standard deviation, mode, maximum, minimum, skewness and kurtosis of the vibration data respectively; F8-F14 are the mean, standard deviation, mode, maximum, minimum, skewness, and kurtosis of the rotational speed data.
Step three, establishing an open set incremental model;
an incremental scene: during training, an initial model M0 is generated from K initial classes, wherein K is the number of classes of a known class, and K =2 in the embodiment through an extreme value theory (EVM); when a batch of new samples come, classifying the samples of the test set, identifying K classes, and rejecting the new classes as unknown; when a batch of new samples arrive, after the new classes are marked as (K + 1) to (K + S), S is the number of classes learned each time, in this embodiment S =1, incremental learning is performed, the initial model M0 is updated to adapt to the new samples of the existing classes and the new classes, the test samples are classified into classes 1 to (K + S), and the new classes are rejected into "unknown classes"; thus, unknown classes can be continually identified and added.
An open set model: for the central limit theorem, no matter what distribution a random variable X obeys, the mean value of a batch of random variables sampled each time obeys normal distribution; similarly, the extremum theory means that no matter what distribution a random variable X obeys, the maximum value of a batch of random variables sampled each time obeys the extremum distribution F; let z = -m ij Wherein m is ij Is of class A i To class A j Half of the closest distance of all sample points, so the minimum half distance is considered to obey the weber distribution; let y i Is the ith row of data in Y, for sample Y i The probability density function for x of a class may be derived based on a Weibull distribution, as shown in equation 1:
wherein, | | y i -x is x and sample y i Distance of (k) i And λ i Weibull shape and scale parameters, respectively; k is a radical of formula i And λ i Is obtained by fitting to a minimum distance; after obtaining the parameters of all samples, the probability that x does not exceed margin can be expressed as:
ψ i ＝1-F(||x-y i ||；k i ,λ i ) (2)
where F is the distribution function of F, test sample x and class C l The associated probability is:
thus, defining a decision function for classifying the K known classes and identifying the "unknown" new class is as follows, with the output label formula:
where y is the output label, δ is the probability rejection threshold, and K is the number of classes of known classes.
In this embodiment, a class sample is gradually added to simulate an incremental open set scene;
the specific operation is as follows: 1) Selecting 2 types of samples in the training samples for training, generating a model M0, and identifying the health conditions of 5 types of bearings of the test samples (the other 3 types are identified as unknown); 2) Continuing to train the third type sample of the training sample and the model MO to generate a model M1, wherein the health condition of the 3 types of bearings in the test sample can be identified, and the rest 2 types are identified as unknown; 3) And by analogy, continuously adding new data, generating a model, and identifying the data of the test sample.
The experiments are shown in tables 2-5, which are incremental experimental results:
table 2 trains H, ORD data, generating model M0; identifying the health states of five types of bearings in a test sample, and finding out that the health states of other three types can be classified into H and ORD by mistake under the model; the IRD data is trained on the basis of the M0 model to obtain a model M1, and table 3 shows that H, ORD and IRD can be basically classified in the model, and the rest two health states can be well distinguished from the model. The model M2 is obtained by training BD data on the basis of the M1 model, and Table 4 shows that H, ORD, IRD and BD can be basically classified in the model M2, the bearing state CD which is not in training can be well distinguished from the four classes, and the visible model can distinguish unknown classes without losing the characteristic information of the known classes. The CD data is trained on the basis of the M2 model to obtain a model M3, and the health states of five types of bearings, namely H, ORD, IRD, BD and CD, can be basically classified in the model as can be seen in the table 5.
TABLE 2H, model M0 classification results of ORD data training
TABLE 3 Classification results of model M1 trained by adding IRD data on the basis of M0 model
TABLE 4 classification result of model M2 trained by adding BD data on the basis of model M1
TABLE 5 Classification results of model M3 trained with CD data on the basis of M2 model
As shown in fig. 3, the method is a complete process of increment, and it can be seen that the F1 evaluation index continuously rises with the continuous increase of class data; in this embodiment, the present invention is compared with the document "Order-Based Identification of Bearing Defects under Variable Speed specification" (abbreviated as "reference 1"), and the results are shown in table 6 and table 7, wherein the BD and CD accuracy of the present invention is higher than the results of reference 1, ord, IRD, BD and CD recall rate is higher than that of reference 1; as shown in table 8, the accuracy of the method proposed in this example is 81.3% higher than 79% of comparative document 1.
TABLE 6 comparison of the precision ratio of the method of the present invention with that of reference 1
TABLE 7 comparison of the recall ratio of the inventive method with reference 1
TABLE 8 comparison of the overall accuracy of the method of the invention with that of reference 1
In light of the foregoing description of the preferred embodiment of the present invention, many modifications and variations will be apparent to those skilled in the art without departing from the spirit and scope of the invention. The technical scope of the present invention is not limited to the content of the specification, and must be determined according to the scope of the claims.
Claims (5)
1. A bearing state monitoring method based on an open set incremental recognition model of signals is characterized by comprising the following steps:
step one, preprocessing collected bearing data and extracting characteristics;
step two, adopting a PCA algorithm to perform dimensionality reduction on the feature data processed in the step one according to the feature values to obtain a dimensionality-reduced feature component;
and step three, establishing an open set incremental model, identifying the state of the bearing, and detecting abnormal data.
2. The method for monitoring the condition of a bearing based on the open-set incremental signal recognition model according to claim 1, wherein the method comprises the following steps: the preprocessing is to divide the bearing vibration data and the rotating speed data in a 2-second window.
3. The method for monitoring the condition of a bearing based on the open-set incremental signal recognition model according to claim 1, wherein the method comprises the following steps: the feature extraction comprises the steps of carrying out mean value, standard deviation, mode, maximum value, minimum value, skewness and kurtosis on the segmented data.
4. The method for monitoring the condition of the bearing based on the open set incremental recognition model of the signal according to claim 1, wherein the second step specifically comprises the following steps:
s21, setting n rows of d-dimensional data, and forming an n row and d column matrix X by the processed characteristic data according to columns; carrying out zero equalization on each column of X;
s22, solving an eigenvalue of the covariance matrix and a corresponding eigenvector; arranging the eigenvectors into a matrix according to the corresponding eigenvalues from large to small in rows, and taking the first k rows to form a matrix P; wherein, Y = PX is data after dimension reduction to k dimensions.
5. The method for monitoring the condition of the bearing based on the open set incremental recognition model of the signal according to claim 1, wherein the third step specifically comprises the following steps:
firstly, training starts from K initial classes, wherein K is the number of classes of known classes, an initial model is generated through an extreme value theory, bearing state data of a test sample are classified into K classes, and data of the other classes are effectively rejected into unknown classes;
secondly, learning a new class, updating the initial model, classifying the bearing state data of the test sample into K +1 classes by the updated initial model, and effectively rejecting the data of the rest classes which are not learned into unknown classes;
and finally, continuously adding unknown class data and identifying.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211323575.8A CN115687998A (en) | 2022-10-27 | 2022-10-27 | Bearing state monitoring method based on signal open set incremental recognition model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211323575.8A CN115687998A (en) | 2022-10-27 | 2022-10-27 | Bearing state monitoring method based on signal open set incremental recognition model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115687998A true CN115687998A (en) | 2023-02-03 |
Family
ID=85100254
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211323575.8A Pending CN115687998A (en) | 2022-10-27 | 2022-10-27 | Bearing state monitoring method based on signal open set incremental recognition model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115687998A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117150265A (en) * | 2023-08-17 | 2023-12-01 | 中国人民解放军陆军工程大学 | Robust radio frequency signal open set identification method under low signal-to-noise ratio condition |
-
2022
- 2022-10-27 CN CN202211323575.8A patent/CN115687998A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117150265A (en) * | 2023-08-17 | 2023-12-01 | 中国人民解放军陆军工程大学 | Robust radio frequency signal open set identification method under low signal-to-noise ratio condition |
| CN117150265B (en) * | 2023-08-17 | 2024-05-17 | 中国人民解放军陆军工程大学 | Robust radio frequency signal open set identification method under low signal-to-noise ratio condition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110132598B (en) | Fault noise diagnosis algorithm for rolling bearing of rotating equipment | |
| CN109514349B (en) | Tool wear state monitoring method based on vibration signal and Stacking integrated model | |
| Carino et al. | Fault detection and identification methodology under an incremental learning framework applied to industrial machinery | |
| Yang et al. | Bearing fault automatic classification based on deep learning | |
| CN113609569B (en) | Distinguishing type generalized zero sample learning fault diagnosis method | |
| CN115687998A (en) | Bearing state monitoring method based on signal open set incremental recognition model | |
| CN116304820A (en) | Bearing fault type prediction method and system based on multi-source domain transfer learning | |
| CN109434562A (en) | Milling cutter state of wear recognition methods based on partition clustering | |
| CN114462475A (en) | Unsupervised machine abnormal sound detection method and unsupervised machine abnormal sound detection device based on single classification algorithm | |
| Xu et al. | Wear particle classification using genetic programming evolved features | |
| Azamfar et al. | Detection and diagnosis of bottle capping failures based on motor current signature analysis | |
| CN115496108A (en) | Fault monitoring method and system based on manifold learning and big data analysis | |
| Maeda et al. | Method for automatically recognizing various operation statuses of legacy machines | |
| Zhang et al. | An image processing approach to machine fault diagnosis based on visual words representation | |
| CN110956331A (en) | Method, system and device for predicting operation state of digital factory | |
| Wagner et al. | Machine condition monitoring and fault diagnostics with imbalanced data sets based on the KDD process | |
| CN113050602A (en) | Industrial process fault method based on robust semi-supervised discriminant analysis | |
| Decker et al. | Does your model think like an engineer? explainable ai for bearing fault detection with deep learning | |
| CN112860183A (en) | Multisource distillation-migration mechanical fault intelligent diagnosis method based on high-order moment matching | |
| Zhou et al. | Degradation State Recognition of Rolling Bearing Based on K‐Means and CNN Algorithm | |
| CN113424120A (en) | Method for determining a property of a machine, in particular a machine tool, without the need for metrological capture of the property | |
| Feng et al. | Temporal local correntropy representation for fault diagnosis of machines | |
| CN116204825A (en) | Production line equipment fault detection method based on data driving | |
| CN115758086A (en) | Method, device and equipment for detecting faults of cigarette cut-tobacco drier and readable storage medium | |
| CN112884027B (en) | Cutting process real-time state monitoring method and device based on pattern recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |